Foobar8568

AIPocalypse.

> If they keep doing it they'll get a letter from the school and probably a police visit.

It depends of the town/location and region. In our town, police had to be involved several times due to different issues with creeps targeting 12yo or less.

Edit:And I forgot, as it's Switzerland, police moved only because parents were accompanying their kids and almost done justice by themselves. Protecting creeps is a national sport here too.

Where is my Gemma 5?

I post once with a fake/hidden email address to be contacted. I received thousands of DocuSign emails...

Take a look at the most viewed videos... It's gross.

Basically 90% are just crap done by computers (pre AI), targeting preschoolers.

Memory used : 38GB, and I haven't even started a LLM nor podman, I always fight with memory when using LLM on my mac with 48gb.

And I don't remember to have been able to have pushed to 200k context Qwen 3.6. 3.8 is running on my RTX 5090.

I know companies that are using github, even using public repo, and request their teams to not use SOTA models, but are ok with local models. Just stupid policy.

Well... Everything new is old "A vector space model for automatic indexing" 1975 - https://dl.acm.org/doi/10.1145/361219.361220

Companies are avoid risks, and at this stage, sometimes, I feel that all cloud providers are just buying RAM that they would have bought anyway. Now it's ensure that no company will be willing to pay x digits just for cards. One of my clients is stuck in "we are doing things on premise but we are too cheap to spend a few 100k in cards but we don't want to go on clouds.

7) can't use reranker with it, asked by people for one year or two... 8) ...

I could stand Opus 4.6-4.8, I was impressed by the initial fable model. Codex 5.6 sol xhigh feels like the initial release of fable. Qwen 3.8 27b feels like using haiku or sonnet (I quickly stopped trying them).

I had $310 in (free) credit that I used on fable, and I still had a part of the $200 subscription at that time. You know, subscriptions don't end the moment you click on cancel.

Opus 5 in xhigh can't do basic math as well. They dumbed it down to a point where I just cancelled my subscription yesterday. I used to be a $200 subscriber, dropped to $20 after the fable shenanigans, and use it only when I have no usage left with Codex.

/on The prose is load-bearing unbearable — every sentence feels like it was engineered to sound profound rather than to be read.

Or that people don't cheat at work. Or that everyone write excellent code. Or all the code is secured.

I bought a prebuilt with a 5090 a little over than one month ago, whole system was on sales.

The cost of the prebuilt is now lower than today 5090 price.

Looks like slope benchmarks and results, and as usually, people are mixing MTP numbers with non MTP numbers. Or just 100 token input benchmarks. Or just failed ones as actual measures.

https://github.com/Neroued/ninfer/blob/master/docs/performan...

Category MTP3 stochastic sampler DFlash stochastic sampler DFlash greedy Code 1/15 natural stops; 0/15 prompt-complete 2/15 natural stops; 0/15 prompt-complete 0/15 natural stops Story 9/15 natural stops; the nine Chinese outputs pass requested division and minimum length 8/15 natural stops; the eight Chinese outputs pass requested division and minimum length 10/15 natural stops; five Chinese dialogue outputs are under length Translation 15/15 natural stops; 15/15 pass structural checks 15/15 natural stops; 15/15 pass structural checks 15/15 natural stops; 15/15 pass structural checks Structured 0/15 satisfy the requested complete record/script contract 0/15 satisfy the requested complete record/script contract 0/15 satisfy the requested complete record/script contract

And on my "own" "quick" benchmark, it's slower than vllm.

Which model/quant/command line did you use? I can barely get 100 token/secs and for sure clearly not a full context. With vllm, I am limited to 130k tokens with vllm + nvfp4.

What IP? Everything can be duplicated within a 1week to a month...

Knowledge work has changed...But it doesn't solve company internal governance nor politics. When it takes anywhere between 2 weeks and 3months to do anything ( including approvals for non prod access) in most organizations...Code has never been the problem.

Considering the clusterfuck that is opus 5 or even fable, if Qwen 27B is trully better than Opus 4.7 Max, I will rejoice.

I didn't like to use GPT for agentic coding, review yes, but with Opus 5, well I really can't stand anything of that model. I feel that sol xhigh is even better than fable.

Mea-culpa, dry-multiplier generated crap and even more so on Gemma4.

Gemma4 was released a few weeks ago? The problems are still there. Today I started using another "provider" and the problems disappeared. Thanks but no.

I had several issues with unsloth gguf, even for models released a few months back like gemma 4, I have 0 confidence in their models, at this stage, I feel several uncensored are more reliable.

Right now, sol-xhigh is my favorite model. I feel that Opus 5 is dumber than 4.8. Fable is too expensive to do anything (limit of $50, started a prompt at $25, ended up at $75, is bullshit, but at least it's "free credits").

DeepSeek is okay for random API-based stuff, as it's cheap.

Local open models running on a 5090 are hit or miss. I feel that most GGUFs/quants are awful...

And last time I have checked, you still can't run rerankers with it, yet you can download the models. See issue 3368, 2years old now.

They didn't care about openmodels for a long while for some reasons... Now it hits their bottom line, mindshare and shareholders.

I'd believe that the 90s on food industry was way worse than now in some ways, almost no control I suppose, already at least a decade of creating shareholders values etc.

What I find the most amusing is ... "AI code is crap". I think I heard enough time "this codebase is shit, we need a full rewrite" to know that no one agrees what is quality code.

Suddenly, everyone are expert writing the most elegant, clean and without bug code.

I feel that Claude converge to more claudism while Chatgpt sounds more natural.

Both sucks in French but at least chatgpt prose is readable while Claude is awful.